lucid.unique

lucid.unique(a: Tensor, /, sorted: bool = True, axis: int | None = None, return_inverse: bool = False) → Tensor | tuple[Tensor, Tensor]

The unique function returns the unique elements of the input tensor. It optionally sorts the result or preserves the order of first appearance.

Function Signature

def unique(a: Tensor, sorted: bool = True, axis: int | None = None) -> Tensor

Parameters

  • a (Tensor): The input tensor from which to extract unique values.

  • sorted (bool, optional): Whether to return the sorted unique values. Defaults to True.

  • axis (int or None, optional): The axis along which to apply uniqueness. If None, the tensor is flattened. Defaults to None.

Returns

  • Tensor: A tensor of unique elements along the specified axis. If axis=None, the result is 1D. Otherwise, it will have shape determined by unique slices.

    The result does not track gradients and has the same dtype as the input.

Note

This operation is non-differentiable. Gradients will not be propagated through unique().

Tip

The default behavior returns sorted unique values. To preserve the original order of appearance, set sorted=False.

Examples

>>> import lucid
>>> x = lucid.tensor([3, 1, 2, 3, 4, 2])
>>> lucid.unique(x)
Tensor([1, 2, 3, 4], grad=None)

>>> lucid.unique(x, sorted=False)
Tensor([3, 1, 2, 4], grad=None)

>>> y = lucid.tensor([[1, 2], [2, 1], [3, 4]])
>>> lucid.unique(y, axis=0)
Tensor([[1, 2],
        [2, 1],
        [3, 4]], grad=None)